name

Observe idle, controlled load, and cooldown behavior

scope

The lab reads Linux telemetry from /proc and /sys, creates a script and observation files under /opt/lab-classroom/class78/, and runs a short single-process CPU workload. It does not write to sysfs, alter a governor, modify a fan curve, install software, or make a persistent system configuration change.

safety_notes

Do not perform the load phase on hardware with a failed fan, failed pump, known cooling fault, active thermal alarm, unstable power supply, or an unknown vendor temperature limit.
Keep an interactive session available. Press Ctrl-C if temperature rises unexpectedly, fans do not respond, the system becomes unstable, or monitoring reports throttling or an alarm.
The workload uses one Python process for a short interval. It is an observation stimulus, not a benchmark and not a cooling-system qualification test.
Virtual machines and containers may expose incomplete, virtualized, or host-controlled telemetry. Do not infer physical package temperature or wall power from unavailable sensors.
Record ambient conditions separately if a trustworthy room sensor is available. Do not substitute an internal motherboard sensor and label it as room ambient.

steps

step

1

instruction

Create the class workspace with ownership assigned to the current user. This is the only persistent location modified by the lab.

command

sudo install -d -o "$(id -un)" -g "$(id -gn)" -m 0750 /opt/lab-classroom/class78/

step

2

instruction

Inventory the read-only telemetry interfaces. Missing paths are acceptable and indicate that the platform, kernel, permissions, or virtualization layer does not expose that source.

command

printf '%s\n' 'Thermal zones:'; ls -1 /sys/class/thermal/thermal_zone/temp 2>/dev/null || true; printf '%s\n' 'CPU frequency files:'; ls -1 /sys/devices/system/cpu/cpu/cpufreq/scaling_cur_freq 2>/dev/null || true; printf '%s\n' 'Energy counters:'; find /sys/class/powercap -type f -name energy_uj -print 2>/dev/null || true

step

3

instruction

Create the observation script. The CSV uses a long format so that systems with different sensor counts can be compared without inventing absent data.

command

cat > /opt/lab-classroom/class78/run_lab.sh <<'EOF'

!/bin/sh

set -eu
BASE=/opt/lab-classroom/class78
OUT="$BASE/observations.csv"
PID=""
cleanup() {
if [ -n "$PID" ]; then
kill "$PID" 2>/dev/null || true
wait "$PID" 2>/dev/null || true
fi
}
trap cleanup EXIT INT TERM
printf 'timestamp,phase,metric,source,value,unit\n' > "$OUT"
record() {
phase="$1"
ts=$(date --iso-8601=seconds)
load1=$(awk '{print $1}' /proc/loadavg)
printf '%s,%s,load1,/proc/loadavg,%s,runnable_average\n' "$ts" "$phase" "$load1" >> "$OUT"
for z in /sys/class/thermal/thermal_zone; do
[ -r "$z/temp" ] || continue
raw=$(cat "$z/temp")
type=$(cat "$z/type" 2>/dev/null | tr -d '\n' || printf unknown)
value=$(awk -v r="$raw" 'BEGIN { printf "%.3f", r / 1000 }')
printf '%s,%s,thermal_c,%s:%s,%s,C\n' "$ts" "$phase" "$z" "$type" "$value" >> "$OUT"
done
for f in /sys/devices/system/cpu/cpu/cpufreq/scaling_cur_freq; do
[ -r "$f" ] || continue
value=$(cat "$f")
printf '%s,%s,cpu_frequency,%s,%s,kHz\n' "$ts" "$phase" "$f" "$value" >> "$OUT"
done
find /sys/class/powercap -type f -name energy_uj -print 2>/dev/null | while IFS= read -r f; do
[ -r "$f" ] || continue
value=$(cat "$f")
printf '%s,%s,energy_counter,%s,%s,uJ\n' "$ts" "$phase" "$f" "$value" >> "$OUT"
done
}
record idle_start
sleep 10
record idle_end
python3 -c 'import hashlib,time
end=time.monotonic()+10
value=b"class78"
while time.monotonic()<end:
value=hashlib.sha256(value).digest()' &
PID=$!
i=1
while [ "$i" -le 10 ]; do
record "load_$i"
sleep 1
i=$((i + 1))
done
wait "$PID" || true
PID=""
sleep 15
record cooldown
printf 'Completed. Review %s\n' "$OUT"
EOF
chmod 0750 /opt/lab-classroom/class78/run_lab.sh

step

4

instruction

Record context needed to interpret the observations. Add only factual details known for the current system.

command

{ date --iso-8601=seconds; uname -a; printf 'Virtualization: '; systemd-detect-virt 2>/dev/null || printf 'unknown\n'; printf 'Online CPUs: '; cat /sys/devices/system/cpu/online; printf 'Operator notes: workload state, room conditions, chassis condition, and active power profile should be documented before comparison.\n'; } > /opt/lab-classroom/class78/context.txt

step

5

instruction

Run the observation sequence while watching the host, its fans, and any available out-of-band thermal console.

command

/opt/lab-classroom/class78/run_lab.sh

step

6

instruction

Inspect the collected rows without modifying system policy.

command

sed -n '1,240p' /opt/lab-classroom/class78/observations.csv

step

7

instruction

Summarize which metrics were actually available. A zero count is a valid finding and must not be replaced with an assumed value.

command

awk -F, 'NR>1 {count[$3]++} END {for (m in count) print m, count[m]}' /opt/lab-classroom/class78/observations.csv | sort > /opt/lab-classroom/class78/metric-counts.txt && cat /opt/lab-classroom/class78/metric-counts.txt

analysis_questions

Which thermal zones changed during the load phase, and are their names specific enough to identify a component?
Did frequency observations vary by CPU, and could the samples represent momentary values rather than state residency?
Did the cooldown observation move toward the idle observations, remain elevated, or become unavailable?
If energy counters exist, which power domain does each counter represent, and can counter wraparound occur?
What environmental or workload variables could prevent a fair comparison with a later run?
Would a proposed optimization reduce total energy, merely reduce peak power, or only move heat and work to another device?